A cybersecurity start-up named Cathedral recently alerted artificial intelligence developer Anthropic that its researchers had successfully accessed sensitive internal data and messaging systems within the company, according to a draft report reviewed by multiple sources. Cathedral, founded by former officials from the U.S. Department of Governmental Executive (DOGE), uses AI-driven cybersecurity solutions and provides services to government and corporate clients.

Cathedral’s researchers gained entry by exploiting vulnerabilities in commonly used third-party business software platforms that Anthropic and many other firms rely on for administrative and sales operations. Their investigation lasted less than 12 hours, during which they uncovered paths to internal data that included meeting transcripts, salary information, job offers, and Slack messages discussing Anthropic’s advanced AI models, according to the report. This probing was conducted under a formal program by Anthropic inviting external experts to test system security.

Anthropic confirmed that Cathedral researchers had accessed “a limited set” of its sales, human resources, and IT systems but did not obtain customer interactions with its AI models or proprietary code, often described as “model weights.” The company quickly closed the exposed vulnerabilities and notified federal authorities. An Anthropic spokesperson said Cathedral initially attempted to breach Anthropic’s own software directly over multiple days before moving to exploit third-party business tools. The start-up’s access was flagged by Anthropic’s monitoring systems and terminated the same day.

The incident comes amid intensifying concerns about the security of U.S.-based AI companies, which are viewed as potential targets of foreign intelligence agencies. A former Trump administration official, speaking anonymously, expressed alarm about how swiftly Cathedral accessed sensitive data at a company regarded as a leader in AI technologies, describing the potential national security implications as significant. The official noted that if a nascent company could penetrate Anthropic’s defenses rapidly, adversaries such as China might be capable of similar actions. Additionally, recent events have spotlighted vulnerabilities within leading AI enterprises, including the FBI’s arrest of an OpenAI employee earlier this month over alleged espionage-related activities.

Cathedral is part of an emerging wave of private cybersecurity firms founded by ex-DOGE personnel seeking contracts with federal agencies; three U.S. government bodies currently engage Cathedral for security-related work. Valued at approximately $1.4 billion, the start-up disclosed its findings to Anthropic in early October, following the protocols of a bug bounty program designed to incentivize vulnerability disclosure. Cathedral also informed the vendors of the third-party software implicated in the breach, though the companies involved were not publicly identified for security reasons.

Industry experts reviewing Cathedral’s findings have pointed to the challenges posed by integrating third-party applications into core corporate systems. Alex Stamos, chief information security officer at AI coding company Cognition, described the issue as a widespread problem across sectors and said the report revealed a notable gap in Anthropic’s management of external software relationships.

Anthropic acknowledged that Cathedral’s report overstated the national security risks and criticized the company for not fully adhering to the bug bounty program’s rules, alleging incomplete disclosure of activity within Anthropic systems. Cathedral responded that all data access was kept minimal in line with best practices for security research.

The episode has sparked internal debate at Anthropic, with former employee Jacob Coxon publicly resigning last month, expressing concern over the company’s handling of sensitive AI technologies and its dependence on conventional business software for critical communications.

Cathedral’s co-founder Gavin Kliger previously served as the Pentagon’s chief data officer and is reported to be collaborating with the current administration on scenarios involving adversarial threats related to AI. The firm has secured $160 million in funding, with prominent investors including Marc Andreessen and Shaun Maguire, both known for their political support of former President Donald Trump and prior involvement with DOGE.